CN109242025A - Model iterative correction methods, apparatus and system - Google Patents
Model iterative correction methods, apparatus and system Download PDFInfo
- Publication number
- CN109242025A CN109242025A CN201811079732.9A CN201811079732A CN109242025A CN 109242025 A CN109242025 A CN 109242025A CN 201811079732 A CN201811079732 A CN 201811079732A CN 109242025 A CN109242025 A CN 109242025A
- Authority
- CN
- China
- Prior art keywords
- model
- basic model
- data
- training
- basic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Abstract
The present invention provides a kind of model iterative correction methods, apparatus and system, are related to Model Updating Technique field, and this method is executed by server, this method comprises: receiving the model feedback information that user terminal uploads；Model feedback information includes problem types and problem amendment data；Search pre-stored basic model corresponding with problem types；Training is iterated to basic model based on problem amendment data, with modified basis model.The present invention can preferably lift scheme the problem of solve efficiency.
Description
Technical field
The present invention relates to Model Updating Technique fields, more particularly, to a kind of model iterative correction methods, apparatus and system.
Background technique
With the development of science and technology, the various neural network models based on deep learning are widely used to all trades and professions, it is all
Such as, can identify the Face datection model of face, the target detection model for being capable of detecting when different classes of target in image, can
Identify that the neural network models with various functions such as the vehicle weight identification model of vehicle have been applied to such as safety-security area, have handed over
Logical field etc..
It is usually (all to model requirements side by model provider (scientific & technical corporation for such as, being absorbed in artificial intelligence technology)
Such as, the group of individual or company etc. using model function is needed) basic model is provided, if basis used by model requirements side
There is identifying any problem such as inaccuracy in model, all can Relationship Model provider under line, and give the customer service of model provider
Deng docking people's feedback problem, then associated inner personnel are transferred to be iterated training to model by model provider, with to asking
The model of topic is modified.This problem solving way very complicated, usually takes a long time, inefficiency.
Summary of the invention
In view of this, being able to ascend the purpose of the present invention is to provide a kind of model iterative correction methods, apparatus and system
The problem of model, solves efficiency.
To achieve the goals above, technical solution used in the embodiment of the present invention is as follows:
In a first aspect, the method is executed by server the embodiment of the invention provides a kind of model iterative correction methods,
The described method includes: receiving the model feedback information that user terminal uploads；The model feedback information includes problem types and problem
Correct data；Search pre-stored basic model corresponding with described problem type；Based on described problem amendment data to institute
It states basic model and is iterated training, to correct the basic model.
Further, the step of lookup pre-stored basic model corresponding with described problem type, comprising: determine
The corresponding model function of described problem type；The basic model for providing the model function is determined as and described problem type pair
The basic model answered.
It is further, described that trained step is iterated to the basic model based on described problem amendment data, comprising:
Obtain the existing training data of the basic model；Data are corrected based on the existing training data and described problem, determination is worked as
Preceding training set；Training is iterated to the basic model by the current training set.
Further, the method also includes: obtain stored corresponding with described problem type history amendment data；Institute
The step of stating based on the existing training data and described problem amendment data, determining current training set, comprising: based on described in
There are training data, described problem amendment data and history amendment data, determines current training set.
Further, described that trained step is iterated to the basic model by the current training set, comprising: will
The current training set is divided into the training subset of preset group number, and the corresponding loss function of training subset described in setting every group and
Weight；Based on the corresponding loss function of training subset and weight described in every group, total losses function is determined；Using training described in each group
Subset is iterated training to the basic model, until the functional value of the total losses function converges to preset function value, really
Determining current iteration terminates.
Further, the method also includes: obtain the basic model has evaluation and test data；Has evaluation and test using described
Data and described problem amendment data evaluate and test the basic model after the current iteration, determine the current iteration
After basic model correction result.
Further, it is described using it is described have evaluation and test data and described problem amendment data to the current iteration after
Basic model the step of being evaluated and tested, determining the correction result of the basic model after the current iteration, comprising: use
It is described have evaluation and test data the basic model after the current iteration is evaluated and tested, after generating the current iteration
Basic model performance index value；The basic model after the current iteration is carried out using described problem amendment data
Evaluation and test, the problem of obtaining the basic model after the current iteration test result；According to the performance index value and described
Problem test result determines the correction result of the basic model after the current iteration.
Further, described according to the performance index value and described problem test result, determine that the current iteration terminates
The step of correction result of basic model afterwards, comprising: if described problem test result is consistent with default correct result, and institute
Performance index value is stated higher than preset threshold, determines that the basic model after the current iteration is corrected successfully；If described ask
It inscribes test result and default correct result is inconsistent, or, the performance index value is lower than the preset threshold, determine described current
Basic model after iteration corrects failure.
Further, the method also includes: correction result is fed back into the user terminal.
Further, described the step of correction result is fed back into the user terminal, comprising: if corrected successfully, will correct
Basic model encapsulation afterwards is sent to the user terminal；If amendment failure, amendment failure information is sent to the use
Family end.
Second aspect, the embodiment of the present invention also provide a kind of model iterated revision device, and described device is set to server
Side, described device include: information receiving module, for receiving the model feedback information of user terminal upload；The model feedback letter
Breath includes that problem types and problem correct data；Model searching module, for searching pre-stored and described problem type pair
The basic model answered；Model iteration module, for being iterated training to the basic model based on described problem amendment data,
To correct the basic model.
The third aspect, the embodiment of the invention provides a kind of model iterated revision system, the system comprises: server and
User terminal；The user terminal is used to the model feedback information of user being uploaded to the server；It is stored on the server
Computer program, the computer program execute such as the described in any item methods of first aspect when being run by the server.
Fourth aspect, the embodiment of the invention provides a kind of computer readable storage medium, the computer-readable storage
Computer program is stored on medium, the computer program is executed when being run by processor described in above-mentioned any one of first aspect
Method the step of.
The embodiment of the invention provides a kind of model iterative correction methods, apparatus and system, server can receive user
The model feedback information (including problem types and problem correct data) uploaded is held, the pre-stored and problem class is then searched
The corresponding basic model of type, and then training can be iterated to the basic model based on problem amendment data, to correct the basis
Model.Aforesaid way provided in this embodiment, user can directly pass through user terminal online feedback problem, and directly by server
Training is iterated to basic model for model feedback information, to correct the problem model.It is this to change online to model
Generation amendment in a manner of solving the problem of model problem it is easier directly can effective lift scheme solve efficiency.
Other feature and advantage of the embodiment of the present invention will illustrate in the following description, alternatively, Partial Feature and excellent
Point can deduce from specification or unambiguously determine, or the above-mentioned technology by implementing the embodiment of the present invention can obtain
Know.
To enable the above objects, features and advantages of the present invention to be clearer and more comprehensible, preferred embodiment is cited below particularly, and cooperate
Appended attached drawing, is described in detail below.
Detailed description of the invention
It, below will be to specific in order to illustrate more clearly of the specific embodiment of the invention or technical solution in the prior art
Embodiment or attached drawing needed to be used in the description of the prior art be briefly described, it should be apparent that, it is described below
Attached drawing is some embodiments of the present invention, for those of ordinary skill in the art, before not making the creative labor
It puts, is also possible to obtain other drawings based on these drawings.
Fig. 1 shows the structural schematic diagram of a kind of electronic equipment provided by the embodiment of the present invention；
Fig. 2 shows a kind of model iterative correction methods flow charts provided by the embodiment of the present invention；
Fig. 3 shows a kind of determination method flow diagram of correction result provided by the embodiment of the present invention；
Fig. 4 shows a kind of structural block diagram of model iterated revision device provided by the embodiment of the present invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with attached drawing to the present invention
Technical solution be clearly and completely described, it is clear that described embodiments are some of the embodiments of the present invention, rather than
Whole embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not making creative work premise
Under every other embodiment obtained, shall fall within the protection scope of the present invention.
In view of model provider solves the mode very complicated of model problem, inefficiency, to improve in the prior art
A kind of the problem of this problem, model provided in an embodiment of the present invention solution, apparatus and system, which can be used relevant
Software and or hardware realization below describes to the embodiment of the present invention in detail.
Embodiment one:
Firstly, describing a kind of model iterative correction methods for realizing the embodiment of the present invention, device referring to Fig.1 and being
The exemplary electronic device 100 of system.
The structural schematic diagram of a kind of electronic equipment as shown in Figure 1, electronic equipment 100 include one or more processors
102, one or more storage devices 104, input unit 106, output device 108 and image collecting device 110, these components
It is interconnected by bindiny mechanism's (not shown) of bus system 112 and/or other forms.It should be noted that electronic equipment shown in FIG. 1
100 component and structure be it is illustrative, and not restrictive, as needed, the electronic equipment also can have other
Component and structure.
The processor 102 can use digital signal processor (DSP), field programmable gate array (FPGA), can compile
At least one of journey logic array (PLA) example, in hardware realizes that the processor 102 can be central processing unit
(CPU) or one or more of the processing unit of other forms with data-handling capacity and/or instruction execution capability
Combination, and can control other components in the electronic equipment 100 to execute desired function.
The storage device 104 may include one or more computer program products, and the computer program product can
To include various forms of computer readable storage mediums, such as volatile memory and/or nonvolatile memory.It is described easy
The property lost memory for example may include random access memory (RAM) and/or cache memory (cache) etc..It is described non-
Volatile memory for example may include read-only memory (ROM), hard disk, flash memory etc..In the computer readable storage medium
On can store one or more computer program instructions, processor 102 can run described program instruction, to realize hereafter institute
The client functionality (realized by processor) in the embodiment of the present invention stated and/or other desired functions.In the meter
Can also store various application programs and various data in calculation machine readable storage medium storing program for executing, for example, the application program use and/or
The various data etc. generated.
The input unit 106 can be the device that user is used to input instruction, and may include keyboard, mouse, wheat
One or more of gram wind and touch screen etc..
The output device 108 can export various information (for example, image or sound) to external (for example, user), and
It and may include one or more of display, loudspeaker etc..
Described image acquisition device 110 can shoot the desired image of user (such as photo, video etc.), and will be clapped
The image taken the photograph is stored in the storage device 104 for the use of other components.
Illustratively, for realizing the example of model iterative correction methods according to an embodiment of the present invention, apparatus and system
Electronic equipment may be implemented as the intelligent terminals such as server, user terminal, host computer, computer.
Embodiment two:
A kind of model iterative correction methods are present embodiments provided, this method can be executed by server, when it is implemented, clothes
The quantity of business device can be realized only with a primary server with sets itself, such as, training can be deployed on the primary server
The basic model of platform and a variety of achievable different function for being supplied to user, basic model namely meets user preset requirement
The model of (such as product fundamental importance)；The training data for training pattern can also be stored on the primary server, also
It can store the evaluation and test data etc. for evaluating and testing model.Certainly, this can also be realized using two servers, a server
(can be described as primary server) is only used for deployment training platform and basic model, another server (can be described as storage server) is used
In storage training data, evaluation and test data etc., in a particular application, storage server can also be realized using other storage equipment,
It is not limited herein.
In one embodiment, reference can be made to a kind of model iterative correction methods flow chart shown in Fig. 2, this method packet
It includes:
Step S202 receives the model feedback information that user terminal uploads；Model feedback information includes problem types and problem
Correct data.For ease of understanding, user terminal, problem types and problem amendment data are illustrated below as follows:
The user terminal of the present embodiment meaning can be the user terminals such as computer, mobile phone, and user can be in user terminal
It is upper that model feedback information is uploaded by named web page, it is also possible to upload model feedback letter by the specified APP on user terminal
Breath.Wherein, server and user terminal data communication, the service implement body can receive user by named web page or specified APP
Model feedback information.
The problem type of the present embodiment meaning can be namely problem class relevant to function provided by basic model
Type can classify according to model function itself, such as, the corresponding problem types of a model function.Such as offer property
The basic model of other arbitration functions, corresponding problem types can be Sexual discriminating mistake；The base of Age estimation function is provided
Plinth model, corresponding problem types can be Age estimation mistake；The basic model of face alignment function is provided, it is corresponding
Problem types can be face alignment mistake etc..For each problem types, the basis of an offer corresponding function can be all corresponded to
Model, certainly, multiple problem types are also possible to corresponding one basic model that can provide multiple functions, such as can progressive
The detection of attribute model of the detections such as not/age.
The problem of the present embodiment meaning corrects data, and the amendment data that can be for solving the problem types such as can
To include original image and the expectation testing result of the original image etc. for causing model inspection mistake.Simple examples are as follows:
(lock is such as misread) when problem types are face alignment mistake, and it may include protoplast's face image that problem, which corrects data, and wrong
Misread the facial image of lock；When problem types are that face age/gender judges incorrectly, problem amendment data may include causing
The original image of model inspection mistake and true age/gender；When problem types are not detect face, problem corrects number
According to the location information etc. that may include face in the original image and original image for cause model inspection mistake.In practical applications,
The feedback interface that user can be presented by user terminal uploads problem and corrects data, can provide problem types choosing in feedback interface
Item, image uploading channel etc..By taking face misreads lock as an example, user can upload two groups of human face photo A and B and select permeability
Type is " misreading lock ".
Step S204 searches pre-stored basic model corresponding with problem types.
It is understood that usually corresponding to different model functions, namely corresponding different for different problems type
Basic model；Therefore it needs according to type search the problem of user feedback to the basic model of correlation function is provided, so as to this
Basic model is iterated amendment.Therefore, the corresponding model function of problem types can be determined in a kind of embodiment；Then will
The basic model for providing model function is determined as basic model corresponding with problem types.
Step S206 is iterated training to basic model based on problem amendment data, with modified basis model.
Basic model is iterated in a kind of trained embodiment, can be used as follows based on problem amendment data
Step: (1) the existing training data of basic model is obtained.(2) data are corrected based on existing training data and problem, determined current
Training set.(3) training is iterated to basic model by current training set.
The problem of above-mentioned model provided in an embodiment of the present invention solution, server can receive user terminal upload mould
Type feedback information (including problem types and problem correct data), then searches pre-stored base corresponding with the problem types
Plinth model, and then training can be iterated to the basic model based on problem amendment data, to correct the basic model.This implementation
The aforesaid way that example provides, user can directly pass through user terminal online feedback problem, and directly anti-for model by server
Feedforward information is iterated training to basic model, to correct the problem model.It is this that amendment is iterated to solve to model online
Certainly the mode of model problem is easier directly, can solve efficiency the problem of effective lift scheme.
In order to be preferably modified to model, in the training set for determining model, can also obtain stored
History corresponding with problem types corrects data, is then based on existing training data, problem amendment data and history amendment data,
Determine current training set.History amendment data be user (can for original subscriber, be also possible to other users) upload with this
The problem of the correspondence of problem types corrects data.In the specific implementation, it can also further be obtained according to user situation selection
History corrects data.Such as, historical data can be obtained to delineation of power for public permission and private permission, if user terminal
It is public permission that historical data, which obtains permission, all history corresponding with problem types stored on server can be corrected
Data are all used as retrievable historic training data；If it is private permission that the historical data of user terminal, which obtains permission, can be only
The history corresponding with the problem types that the user terminal (or the user terminal belonged to corporation etc.) was once uploaded
Data are corrected as retrievable historic training data.
Current training set is being determined, when being iterated trained to basic model by current training set, in order to more
Following steps execution can be used in good training for promotion effect in one implementation:
Step 1, current training set is divided into the training subset of preset group number, and sets the corresponding damage of every group of training subset
Lose function and weight.Preset group number can be N group, and N group training subset can be understood as N number of subtask again, in this way
Realize the multitask optimization in iterative process.With simplest N=2 example, one group can be existing training data, and one group can be with
Data are corrected for problem.This mode is compared with the mode that directly problem amendment data are mixed into existing training data, advantage
Be: less problem amendment data will not largely be had training data and be flooded, the less problem amendment of user feedback
Data can directly have an impact basic model.It is therefore preferable that problem can be corrected data in mode and individually be divided into one
Group.It is, of course, also possible to as needed be grouped existing training data again, or new training data etc. is added, thus
Form multiple groups.
Step 2, it is based on the corresponding loss function of every group of training subset and weight, determines total losses function.The present embodiment is also
Former loss function can be changed according to the actual situation or increases loss function, such as by problem amendment data separately as one
Task is finely adjusted model, and problem corrects the corresponding new loss function of data.Assuming that total N group training subset, then it can be right
N number of loss function and N number of weight are answered, synthesis can determine total losses function by weighting scheme.
Step 3, training is iterated to basic model using each group training subset, until the functional value of total losses function is received
It holds back to preset function value, determines that current iteration terminates.Total losses function characterizes the reality output of model and the difference of desired output
It is different, total losses functional value is smaller namely reality output and the difference of desired output it is smaller, the accuracy of model is higher, when total damage
When mistake functional value converges to preset value, the accuracy that can characterize "current" model reaches basic demand.
Further, for the accuracy rate of further assurance model, the above method provided in this embodiment further include: obtain base
Plinth model has evaluation and test data；Data and problem amendment data are evaluated and tested to the basic model after current iteration using having
It is evaluated and tested, determines the correction result of the basic model after current iteration.Wherein, having assessment data can be test number
According to collection, for being input in basic model, performance evaluating is carried out to basic model according to the testing result of basic model output.It is all
Have assessment data as Face datection model is corresponding and can be and carry the face image set of face location label.
It should be noted that the present embodiment can also when use has evaluation and test data and evaluates and tests to the model after iteration
Individually the model after iteration is evaluated and tested using problem amendment data, to judge that whether the model solved that user points out asks
Topic.Specifically, being referred to a kind of determination method flow diagram of correction result shown in Fig. 3 when determining correction result, wrap
Include following steps:
Step S302 evaluates and tests the basic model after current iteration using having evaluation and test data, generates current
The performance index value of basic model after iteration.Performance index value can for precision etc. can characterization model performance value, with
For Face datection, performance index value can also be face recall rate and non-face false detection rate etc..
Step S304 evaluates and tests the basic model after current iteration using problem amendment data, obtains current
The problem of basic model after iteration test result.Herein by taking age detection as an example, also by the initiation basis of user feedback
The original image of model age identification mistake is input to the basic model after current iteration, the basic model pair after judging current iteration
The testing result (that is, problem test result) of original image.
Step S306 determines the basic model after current iteration according to performance index value and problem test result
Correction result.When it is implemented, if problem test result is consistent with default correct result, and performance index value is higher than default threshold
Value, determines that the basic model after current iteration is corrected successfully；If problem test result and default correct result are inconsistent,
Or, performance index value is lower than preset threshold, the basic model amendment failure after current iteration is determined.Specifically, each
When model consigns to user (model requirements side), all there is threshold requirement, it is assumed that it is required that the accuracy rate of model is 98%, solving
When certainly the problem of user feedback, it is also necessary to guarantee the accuracy rate of model not less than 98%；If being once lower than 98%, can not expire
The basic demand of sufficient user.
It is understood that basic model is after each iteration, all evaluated and tested using evaluation and test data, to judge current change
Whether basic model after generation corrects failure.Such as, it if the basic model after current iteration is corrected successfully, can know
The correction result of basic model is successfully.If the basic model after current iteration corrects failure, and model at this time
Energy index has been lower than preset threshold, illustrates that basic property has been unable to satisfy user demand, then can know the amendment knot of basic model
Fruit is failure.
Correction result can also be fed back to user terminal after being trained iteration to basic model by server.If
It corrects successfully, the encapsulation of revised basic model is sent to user terminal；If amendment failure information is sent to by amendment failure
User terminal, to realize real-time client feedback.
In conclusion above-mentioned model iterative correction methods provided in this embodiment, pass through what is be iterated online to model
Mode can solve efficiency the problem of effective lift scheme, help that correction result is fed back to client within a short period of time, from
And preferably the user experience is improved degree.
Embodiment three:
Corresponding to a kind of aforementioned model iterative correction methods, a kind of model iterated revision device is present embodiments provided, it should
Device may be disposed at server side, a kind of structural block diagram of model iterated revision device shown in Figure 4, which includes:
Information receiving module 402, for receiving the model feedback information of user terminal upload；Model feedback information includes problem
Type and problem correct data；
Model searching module 404, for searching pre-stored basic model corresponding with problem types；
Model iteration module 406, for being iterated training to basic model based on problem amendment data, with modified basis
Model.
Above-mentioned model iterated revision device provided in an embodiment of the present invention, server can receive the model of user terminal upload
Feedback information (including problem types and problem correct data), then searches pre-stored basis corresponding with the problem types
Model, and then training can be iterated to the basic model based on problem amendment data, to correct the basic model.The present embodiment
The aforesaid way of offer, user can directly pass through user terminal online feedback problem, and directly be directed to model feedback by server
Information is iterated training to basic model, to correct the problem model.It is this that amendment is iterated to solve to model online
The mode of model problem is easier directly, can solve efficiency the problem of effective lift scheme.
In one embodiment, model searching module is used for: determining the corresponding model function of problem types；Mould will be provided
The basic model of type function is determined as basic model corresponding with problem types.
In one embodiment, model iteration module is used for: obtaining the existing training data of basic model；Based on existing
Training data and problem correct data, determine current training set；Training is iterated to basic model by current training set.
In one embodiment, above-mentioned apparatus further include: historical data obtain module, for obtain it is stored with ask
It inscribes the corresponding history of type and corrects data.Based on this, model iteration module is further used for: based on existing training data, problem
It corrects data and history corrects data, determine current training set.
In a specific embodiment, model iteration module is further used for: current training set is divided into preset group
Several training subsets, and set the corresponding loss function of every group of training subset and weight；Based on the corresponding damage of every group of training subset
Function and weight are lost, determines total losses function；Training is iterated to basic model using each group training subset, until total losses
The functional value of function converges to preset function value, determines that current iteration terminates.
In one embodiment, above-mentioned apparatus further include: evaluation and test data acquisition module and evaluation and test module, wherein evaluation and test
What data acquisition module was used to obtain basic model has evaluation and test data；Evaluation and test module, which is used to use, has evaluation and test data and problem
Amendment data evaluate and test the basic model after current iteration, determine the amendment of the basic model after current iteration
As a result.
In one embodiment, above-mentioned evaluation and test module is further used for: evaluating and testing data to current iteration knot using having
Basic model after beam is evaluated and tested, the performance index value of the basic model after generation current iteration；It is corrected using problem
Data evaluate and test the basic model after current iteration, obtain current iteration after basic model the problem of test
As a result；According to performance index value and problem test result, the correction result of the basic model after current iteration is determined.
In one embodiment, above-mentioned evaluation and test module is further used for: if problem test result and default correct knot
Fruit is consistent, and performance index value is higher than preset threshold, determines that the basic model after current iteration is corrected successfully；If problem
Test result and default correct result are inconsistent, or, performance index value is lower than preset threshold, determine the base after current iteration
The failure of plinth Modifying model.
Above-mentioned apparatus further include: feedback module, for correction result to be fed back to user terminal.
In one embodiment, above-mentioned feedback module is used for: if corrected successfully, revised basic model being encapsulated
It is sent to user terminal；If amendment failure, amendment failure information is sent to user terminal.
The technical effect of device provided by the present embodiment, realization principle and generation is identical with previous embodiment, for letter
It describes, Installation practice part does not refer to place, can refer to corresponding contents in preceding method embodiment.
Example IV:
The embodiment of the invention also provides a kind of model iterated revision system, which includes: server and user terminal；Its
In, user terminal is used to the model feedback information of user being uploaded to server；Computer program, computer are stored on server
Program executes the method such as any one of embodiment two when being run by server.
It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description
Specific work process, can refer to corresponding processes in the foregoing method embodiment, details are not described herein.
Further, the present embodiment additionally provides a kind of computer readable storage medium, the computer readable storage medium
On be stored with computer program, the matching method of above-mentioned article and electronic tag is executed when computer program is run by processor
Step.
The computer program product of model iterative correction methods, apparatus and system provided by the embodiment of the present invention, including
The computer readable storage medium of program code is stored, the instruction that said program code includes can be used for executing previous methods reality
Method described in example is applied, specific implementation can be found in embodiment of the method, and details are not described herein.
It, can be with if the function is realized in the form of SFU software functional unit and when sold or used as an independent product
It is stored in a computer readable storage medium.Based on this understanding, technical solution of the present invention is substantially in other words
The part of the part that contributes to existing technology or the technical solution can be embodied in the form of software products, the meter
Calculation machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be a
People's computer, server or network equipment etc.) it performs all or part of the steps of the method described in the various embodiments of the present invention.
And storage medium above-mentioned includes: that USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), arbitrary access are deposited
The various media that can store program code such as reservoir (RAM, Random Access Memory), magnetic or disk.
Finally, it should be noted that embodiment described above, only a specific embodiment of the invention, to illustrate the present invention
Technical solution, rather than its limitations, scope of protection of the present invention is not limited thereto, although with reference to the foregoing embodiments to this hair
It is bright to be described in detail, those skilled in the art should understand that: anyone skilled in the art
In the technical scope disclosed by the present invention, it can still modify to technical solution documented by previous embodiment or can be light
It is readily conceivable that variation or equivalent replacement of some of the technical features；And these modifications, variation or replacement, do not make
The essence of corresponding technical solution is detached from the spirit and scope of technical solution of the embodiment of the present invention, should all cover in protection of the invention
Within the scope of.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Claims (13)
1. a kind of model iterative correction methods, which is characterized in that the method is executed by server, which comprises
Receive the model feedback information that user terminal uploads；The model feedback information includes problem types and problem amendment data；
Search pre-stored basic model corresponding with described problem type；
Training is iterated to the basic model based on described problem amendment data, to correct the basic model.
2. the method according to claim 1, wherein the lookup is pre-stored corresponding with described problem type
Basic model the step of, comprising:
Determine the corresponding model function of described problem type；
The basic model for providing the model function is determined as basic model corresponding with described problem type.
3. according to the method described in claim 2, it is characterized in that, described correct data to the basic mould based on described problem
Type is iterated trained step, comprising:
Obtain the existing training data of the basic model；
Data are corrected based on the existing training data and described problem, determine current training set；
Training is iterated to the basic model by the current training set.
4. according to the method described in claim 3, it is characterized in that, the method also includes: obtain and stored asked with described
It inscribes the corresponding history of type and corrects data；
It is described that data are corrected based on the existing training data and described problem, the step of determining current training set, comprising:
Data are corrected based on the existing training data, described problem amendment data and the history, determine current training set.
5. according to the method described in claim 3, it is characterized in that, it is described by the current training set to the basic model
It is iterated trained step, comprising:
The current training set is divided into the training subset of preset group number, and the corresponding loss of training subset described in setting every group
Function and weight；
Based on the corresponding loss function of training subset and weight described in every group, total losses function is determined；
Training is iterated to the basic model using training subset described in each group, until the functional value of the total losses function
Preset function value is converged to, determines that current iteration terminates.
6. method according to any one of claims 1 to 5, which is characterized in that the method also includes:
Obtain the basic model has evaluation and test data；
Using it is described have evaluation and test data and described problem amendment data the basic model after current iteration is evaluated and tested,
Determine the correction result of the basic model after the current iteration.
7. according to the method described in claim 6, it is characterized in that, it is described using it is described have evaluation and test data and described problem repair
Correction data evaluates and tests the basic model after the current iteration, determines the basic model after the current iteration
Correction result the step of, comprising:
Have evaluation and test data using described the basic model after the current iteration evaluated and tested, generate it is described it is current repeatedly
The performance index value of basic model after generation；
The basic model after the current iteration is evaluated and tested using described problem amendment data, acquisition is described currently to change
Generation after basic model the problem of test result；
According to the performance index value and described problem test result, repairing for the basic model after the current iteration is determined
Positive result.
8. the method according to the description of claim 7 is characterized in that described test according to the performance index value and described problem
As a result, the step of correction result of basic model after determining the current iteration, comprising:
If described problem test result is consistent with default correct result, and the performance index value is higher than preset threshold, determines
Basic model after the current iteration is corrected successfully；
If described problem test result and default correct result are inconsistent, or, the performance index value is lower than the default threshold
Value determines the basic model amendment failure after the current iteration.
9. the method according to claim 1, wherein the method also includes: correction result is fed back to described
User terminal.
10. according to the method described in claim 9, it is characterized in that, the step that correction result is fed back to the user terminal
Suddenly, comprising:
If corrected successfully, the revised basic model encapsulation is sent to the user terminal；
If amendment failure, amendment failure information is sent to the user terminal.
11. a kind of model iterated revision device, which is characterized in that described device is set to server side, and described device includes:
Information receiving module, for receiving the model feedback information of user terminal upload；The model feedback information includes problem class
Type and problem correct data；
Model searching module, for searching pre-stored basic model corresponding with described problem type；
Model iteration module, for being iterated training to the basic model based on described problem amendment data, to correct
State basic model.
12. a kind of model iterated revision system, which is characterized in that the system comprises: server and user terminal；
The user terminal is used to the model feedback information of user being uploaded to the server；
Computer program is stored on the server, the computer program executes such as right when being run by the server
It is required that 1 to 10 described in any item methods.
13. a kind of computer readable storage medium, computer program, feature are stored on the computer readable storage medium
The step of being, the described in any item methods of the claims 1 to 10 executed when the computer program is run by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811079732.9A CN109242025B (en) | 2018-09-14 | 2018-09-14 | Model iteration correction method, device and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811079732.9A CN109242025B (en) | 2018-09-14 | 2018-09-14 | Model iteration correction method, device and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109242025A true CN109242025A (en) | 2019-01-18 |
| CN109242025B CN109242025B (en) | 2021-05-04 |
Family
ID=65058524
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811079732.9A Active CN109242025B (en) | 2018-09-14 | 2018-09-14 | Model iteration correction method, device and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109242025B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110942004A (en) * | 2019-11-20 | 2020-03-31 | 深圳追一科技有限公司 | Handwriting recognition method and device based on neural network model and electronic equipment |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7222127B1 (en) * | 2003-11-14 | 2007-05-22 | Google Inc. | Large scale machine learning systems and methods |
| CN104680046A (en) * | 2013-11-29 | 2015-06-03 | 华为技术有限公司 | User activity recognition method and device |
| CN105912500A (en) * | 2016-03-30 | 2016-08-31 | 百度在线网络技术（北京）有限公司 | Machine learning model generation method and machine learning model generation device |
| CN107766940A (en) * | 2017-11-20 | 2018-03-06 | 北京百度网讯科技有限公司 | Method and apparatus for generation model |
| CN108197664A (en) * | 2018-01-24 | 2018-06-22 | 北京墨丘科技有限公司 | Model acquisition methods, device, electronic equipment and computer readable storage medium |
-
2018
- 2018-09-14 CN CN201811079732.9A patent/CN109242025B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7222127B1 (en) * | 2003-11-14 | 2007-05-22 | Google Inc. | Large scale machine learning systems and methods |
| CN104680046A (en) * | 2013-11-29 | 2015-06-03 | 华为技术有限公司 | User activity recognition method and device |
| CN105912500A (en) * | 2016-03-30 | 2016-08-31 | 百度在线网络技术（北京）有限公司 | Machine learning model generation method and machine learning model generation device |
| CN107766940A (en) * | 2017-11-20 | 2018-03-06 | 北京百度网讯科技有限公司 | Method and apparatus for generation model |
| CN108197664A (en) * | 2018-01-24 | 2018-06-22 | 北京墨丘科技有限公司 | Model acquisition methods, device, electronic equipment and computer readable storage medium |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110942004A (en) * | 2019-11-20 | 2020-03-31 | 深圳追一科技有限公司 | Handwriting recognition method and device based on neural network model and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109242025B (en) | 2021-05-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109299344A (en) | The generation method of order models, the sort method of search result, device and equipment | |
| CN112632385A (en) | Course recommendation method and device, computer equipment and medium | |
| CN109241711A (en) | User behavior recognition method and device based on prediction model | |
| CN109376615A (en) | For promoting the method, apparatus and storage medium of deep learning neural network forecast performance | |
| CN110766080B (en) | Method, device and equipment for determining labeled sample and storage medium | |
| CN110163647A (en) | A kind of data processing method and device | |
| CN109241709A (en) | User behavior recognition method and device based on the verifying of sliding block identifying code | |
| JP2022512065A (en) | Image classification model training method, image processing method and equipment | |
| CN109600336A (en) | Store equipment, identifying code application method and device | |
| WO2014053192A1 (en) | Method and apparatus for ranking users within a network | |
| CN111291773A (en) | Feature identification method and device | |
| CN110390107A (en) | Hereafter relationship detection method, device and computer equipment based on artificial intelligence | |
| CN110232128A (en) | Topic file classification method and device | |
| CN109743286A (en) | A kind of IP type mark method and apparatus based on figure convolutional neural networks | |
| CN107291774B (en) | Error sample identification method and device | |
| CN115187772A (en) | Training method, device and equipment of target detection network and target detection method, device and equipment | |
| CN114270365A (en) | Elastic centroid based clustering | |
| EP4273750A1 (en) | Data processing method and apparatus, computing device, and test simplification device | |
| CN109685104B (en) | Determination method and device for recognition model | |
| CN115081613A (en) | Method and device for generating deep learning model, electronic equipment and storage medium | |
| CN106897282A (en) | The sorting technique and equipment of a kind of customer group | |
| CN113190444B (en) | Test method, test device and storage medium | |
| CN110581856A (en) | malicious code detection method and system | |
| CN107729947A (en) | A kind of Face datection model training method, device and medium | |
| CN105162931B (en) | The sorting technique and device of a kind of communicating number |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |